› Publications
› Software

Cite Details

David Pérez, Brendt Wohlberg, Thomas Alan Lovell, Michael Shoemaker and Riccardo Bevilacqua, "Orbit-Centered Atmospheric Density Prediction using Artificial Neural Networks", Acta Astronautica, vol. 98, doi:10.1016/j.actaastro.2014.01.007, pp. 9--23, May 2014


At low Earth orbits, drag force is a significant source of error for propagating the motion of spacecraft. The main factor driving changes on the drag force is the neutral density. Global atmospheric models provide estimates for the density which are significantly affected by bias due to misrepresentations of the underlying physics and limitations on the statistical models. In this work a localized predictor based on artificial neural networks is presented. Localized refers to the focus being on a specific orbit, rather than a global prediction. The predictor uses density measurements or estimates on a given orbit and a set of proxies for solar and geomagnetic activities to predict the value of the density along the future orbit of the spacecraft. The performance of the localized predictor is studied for different neural network structures, testing periods of high and low solar and geomagnetic activities and different prediction windows. Comparison with previously developed methods show substantial benefits in using artificial neural networks, both in prediction accuracy and in the potential for spacecraft onboard implementation. In fact, the proposed neural networks are computationally efficient and would be straightforward to integrate into onboard software.

BibTeX Entry

author = {David P\'{e}rez and Brendt Wohlberg and Thomas Alan Lovell and Michael Shoemaker and Riccardo Bevilacqua},
title = {Orbit-Centered Atmospheric Density Prediction using Artificial Neural Networks},
year = {2014},
month = May,
urlpdf = {},
journal = {Acta Astronautica},
volume = {98},
doi = {10.1016/j.actaastro.2014.01.007},
pages = {9--23}